Knowledge Acquisition, Modification And Reasoning In Expert System Based On Rough Set Theory

With the development of the application system of the artificial intelligence and the knowledge engineering, the expert system is an extremely active branch domain. It was successfully applied in many domains. With the enhancement of demand, many questions such as the knowledge acquisition and the uncertain knowledge reasoning are exposed.Rough set theory that was put forward by Pole Z. Pawlak in 1982 is a new data analysis theory of analyzing and dealing with uncertain and incomplete data. It makes use of the equivalence relations to measure the indetermination degree of knowledge and it doesn't need any knowledge outside of the data which needs to be processed. Therefore the error caused by subjective appraisal can be avoided. So the studies of the uncertain knowledge acquisition, modification and reasoning based on rough set theory have widespread application prospect.The main works of this paper are that a basic model structure of an expert system based on rough set theory is given, and the knowledge acquisition mechanism, the knowledge modification mechanism and the knowledge reasoning mechanism based on the model are researched in detail. In the study of the knowledge acquisition mechanism, several traditional knowledge acquisition algorithms are introduced, and several improved knowledge acquisition algorithms are proposed to overcome the traditional algorithms'shortcoming. In the study of the knowledge modification mechanism, a new algorithm for incremental learning is proposed, and the situation of decrement learning is discussed. In the study of the knowledge reasoning mechanism, several common methods for uncertain reasoning are introduced simply, an uncertain reasoning method considering subjective factors is given and a reasoning method about incomplete knowledge is discussed.The main innovation selects in the paper are as follows:(1) A basic model structure of an expert system based on rough set theory is given by improving the traditional model structure ( like chart 1.3.1 ).(2) Traditional algorithms for reduction of attributes are improved. The weight coefficient and the subjective factor are considered in the process of reduction.(3) The definition method of traditional reduction of rules is expanded, and the definition method is suitable for inconsistent rules. An algorithm for reduction of rules based on the character of Apriori and the support is proposed.(4) When a new instance is added to the decision table, the minimal set of rules will be changed correspondingly. In the paper, the situations of all new instances are classified according the relationship between the new instance and the old minimal set of rules, and a new algorithm for incremental learning based on this classification is proposed.(5) The situation that the minimal set of rules will change when an instance is deleted from the knowledge base is researched.(6) The importance of attributes in the new object is considered in the process of the knowledge reasoning, and an uncertain reasoning method considering subjective factors is given.(7) The traditional definition of the match set of the new object is expanded in the process of incomplete knowledge reasoning. Based on this, a method for the incomplete knowledge reasoning is researched.Some traditional algorithms for the knowledge acquisition, modification and reasoning are improved in the paper, but there are some shortcomings in some aspects. For example: the algorithms'efficiency of the knowledge acquisition in the paper is still low, and there are errors. An effective evaluation criterion in the question of measuring the relationship between simple and integrity is required. Although the necessity of recalculating is reduced greatly in the improved algorithm for incremental learning, the efficiency of the algorithm is lower when the reduction of condition attributes is changed after adding the new instance. In the uncertain reasoning considering subjective factors, the proportion of subjective factors to objective factors is given by the user. But how to give the proportion objectively will be researched in the future.